Re: Asynchronous MergeAppend

On Mon, Oct 5, 2026 at 6:10 PM Etsuro Fujita <etsuro(dot)fujita(at)gmail(dot)com> wrote:
> On Mon, Oct 5, 2026 at 2:36 AM Alexander Korotkov <aekorotkov(at)gmail(dot)com> wrote:
> > Are you still going to review this patch?

> I'm planning to
> provide my initial comments by the end of next week, hopefully.

I've returned to this topic. Here is an overall comment on the patch:

I think you are creating the patch based on the idea that async
MergeAppend should work the same way as async Append: specifically, 1)
introduce common functionality for that by making really
large/invasive changes to the planner/executor around
Append/MergeAppend, and then 2) (re-)implement async
Append/MergeAppend on top of it. I'm very sorry to say this so late,
but I think you are misguided from the beginning. Append/MergeAppend
look similar, but they are actually different at least from the
perspective of async execution (one such example is the ReScan
handling we've discussed before). So I think we should implement
async MergeAppend in a more optimized way, separately from async
Append. To put it another way, I think it's *too* early to consider
refactoring like #1. If both async Append/MergeAppend were (almost)
finished, we could do so, but they would actually be far from
finished, so doing the refactoring at this point would just make the
future maintenance/development of them hard. Again, they are actually
different. I think it's a good idea to do refactoring to share
between them some code like the guts of ExecAppendAsyncEventWait() in
async Append, though. I've already started reworking on Alexander
Pyhalov's original patch for that. I will post a reworked version as
soon as possible.

Best regards,
Etsuro Fujita